 Good to be here. How's everybody doing? My name is Siraj Raval and Hello World. That's my intro on YouTube, where I inspire and educate lots of developers to build AI. Today, my talk is going to be about decentralized artificial intelligence, a.k.a. what happens when we take the two hottest technologies of today, AI, and blockchain and put them together, right? Just sprinkle everything on top of each other, right? So let's get right on into this. We've got a lot to cover. So machine learning, AI, this has been around since the 50s. It's not a new technology. It's a pretty old technology. But basically, the idea has always been the same. We have some data sets, right? Some input data set. And we have some objective. We have something that we want the AI to do. In this case, classify a car. Is it a car or is it not a car? So what we do is we feed this AI a bunch of images of cars. So think of it like an Excel spreadsheet with two columns. In column one, you just have pictures of cars. In column two, you just have labels. Car, car, car, car, car, car, car, car, car, car, right? And so the idea is that the AI learns the mapping, the relationship between the input data and the output labels. There is some relationship between everything. There is a function in life for everything. But that's a bit of a tangent. The idea, though, is that the human would have to extract all the features of what it means to be a car, right? This is kind of what a car looks like. This is the shape of a car. This is what the exhaust pipe looks like. This was a very tiring process. But this is what humans had to do. We'd extract features by hand. And so what happened is, once we've extracted those features, we would feed it to some kind of machine learning model. And there's a lot out there. Eventually, it would learn the mapping and then give it a new car picture. And it would know, hey, this is a car. But a few years ago, there was a scientist who said, let me take one specific machine learning model called a neural network. And sure, drop the beat, right? Drop the mic. And feed it lots of data and lots of computing power and add a lot of layers, deep layers, AKA deep learning. That's what we call it. What happened is, when we did this, it started outperforming every other machine learning model. And now, deep learning is the hottest topic in machine learning. It is outperforming everything else, self-driving cars, drug discovery. Everything hot that you're hearing about AI is coming from deep learning, OK? So that was my one slide overview of AI. So now, blockchain, right? So a couple years ago, there was a programmer who no one knows, named Satoshi Nakamoto, who released a paper on a cryptography mailing list detailing a system called Bitcoin that allows two people to transmit value online without needing a third party, namely a bank. So what happens is, instead of a bank being the third party, there are a group of people called miners. And anybody can become a miner. You just need a laptop, right? Anybody can become a miner. And the idea is that when I transmit value to you, from me, these miners have to approve this transaction. They have to say, OK, let me check this list of transactions. So every miner has a copy of every transaction that has occurred in the network. And they have to approve whether or not this transaction is valid or not, because they have the list. So you might be thinking, wait a second. Couldn't someone just create a bunch of accounts and say, hey, I'm the majority of the miners, because the majority of the miners have to approve a transaction for it to be added to this list of transactions? Well, no, because Satoshi said every single miner has to prove that they have solved some random mathematical problem. It's called the proof of work algorithm. And that means that you have to have more computing power than the 500 fastest supercomputers in the world combined to have the majority of the computing power in the Bitcoin network. And because no one has that much computing power, no one's been able to hack Bitcoin. And that's why it has a over 100 billion market cap, as of yesterday, more than the GDP of some countries and no one's been able to hack it. It's been around for a decade. It's a really powerful technology. Let's keep going here. But the really, really interesting part about Bitcoin is not the actual application, but the underlying data structure of Bitcoin. So this is kind of a rough diagram of what it looks like, very simplified. But the idea is that we have some list of transactions. I'm talking about this list of transactions that all of the miners are holding. And we group these transactions together into little groups that we call blocks. And each block points to the next block. So one block is like, okay, I'm a list of transactions. Here's the next block. And the next block's like, oh, thanks for that. Okay, here's the next block, right? And it becomes this kind of chain of blocks, AKA blockchain, right? So it's a blockchain. That's what we call this data structure. And because of the proof of work algorithm, no one can modify it. It's an un-ultrable, immutable, in computer science terms, shout out to all the programmers out there. It's an immutable data structure that no one owns. And this is a very powerful idea. So I like to call this the yin and the yang of AI and blockchain, right? These two technologies go really well together, but no one's really put them together yet. So this is a very futuristic thing. So the yin is AI. AI is probabilistic. It's all about computing the likelihood that something will happen, the prediction of the future using what it's learned. And AI is always changing. It's updating its weights. It's learning about the future. And an AI is a set of algorithms that guess at reality. The yang part are blockchains. Blockchains are not probabilistic. They are deterministic. You know exactly what's gonna happen, right? They are permanent. They are unchangeable. And they are algorithms and cryptography to record reality. So these two things can go really well together if you break them down into what they're really good at. So we live in a world where we are all using these services, right? And these services collect and monetize our data. That's the business model. But this is a problem because our data is the most valuable thing we have. And it will become more valuable as everything is slowly automated. Your data will be the only agency that you have. It will be your most valuable asset. So you need to own it. But right now we're giving it away for free in return for a free service. Ideally, these services would be paying us for our data, right? But that's just not the case right now. We need to redo all of these apps that we have in Web 2.0. Everything from eBay to Uber to Lyft, all of it, Facebook, right? Morally speaking, this is a terrible idea to let some humans control our data. In the case of Facebook, they modified the newsfeed to affect large swaths of the populace. And the 2016 political elections, for example, in the US, it's just very corrupt. It's just not a good system. And not even talking about it morally speaking, let's talk about it technically speaking. Centralized systems are not good. Take the case of self-driving car network. Let's say we've got three self-driving cars at an intersection, okay? Three of them at an intersection. And they're gonna decide who has the right of way. But if there was one central server that they all had to talk to a million miles or kilometers away, then it would take forever. And there would probably be an accident. Ideally, these cars could talk to each other without needing a third party or a central service or a server. So both technically speaking and morally speaking, creating what are called decentralized applications is a really good idea. And it's what we call as developers, Web 3.0. And that's what's happening now. If you're looking for something to disrupt, you take blockchain, you take AI, you combine them together, and you create apps that have never before been possible. And that's the most exciting thing these days. So this leads us to this idea. It's a very abstract idea, but it's starting to happen of what are called decentralized autonomous organizations or DAOs for short. You can also call them DAPS, Decentralized Apps. But the idea is that these are apps that are run and maintained by the community that uses them, right? So think of Uber, for example. When I'm a rider and I'm trying to find a driver, I have to use a third party, namely Uber, to find that driver. And so Uber takes a cut of this and they have a bunch of malpractice as well that we all know of through all of their scandals. But ideally, I wouldn't have to trust Uber. I could just find my driver. He would get paid more. I would have to pay less because there's no third party. So it would be better for both of us, right? We would all earn some equity because we are all a part of the network. It's not just about earning money. We'd also earn equity. And we would be able to decide on the features of the app. So it would be a more democratic process, right? It's kind of like a community-owned grocery store, for example. Instead of one central person owning it, everybody who is a part of the network would also be a participant in the infrastructure of that network as well. So there's a lot of good that can come from this. There's a lot of good that can come from this. And that's what we're going to start to see as we move towards Web 3.0. If you're looking for a really hot idea, something really cool to do, this is it. Blockchain, AI, combine them together, use this immutable ledger, and have an AI live on the blockchain, or at least speak to the blockchain. And data is being pointed to to some distributed hash table or some kind of decentralized storage source. And then there's a whole token model. I've only got two and a half minutes, so I can't explain everything here. But there's so much. There's so much. That's what I'm trying to say. But it's not all good. There's also bad. I can't talk about the good without talking about the bad, right? So imagine this. What you're seeing here is a picture of art that was completely generated by AI. There is no human in the loop. So imagine an AI that lives on a blockchain and its job is to just generate art. So what it does, it does this. It generates some art. It then sells it on some marketplace. It earns some money. It uses that money to feed itself more computing power. It uses that computing power to then generate more art. It just keeps amassing wealth. And no one can shut it down because it lives on a blockchain, right? No one can shut it down. You'd have to have more computing power than the 500 fastest supercomputers in the world combined to shut it down. Eventually it becomes the wealthiest entity on the planet. Right? So that's kind of scary if you think about it. It's an unkillable AI and it lives on a blockchain. So how do we stop this? It's our responsibility as developers, as people, people who don't even program to educate ourselves on how this intelligence works. And only then will we be able to stop any kind of malpractice, any kind of bad scenario. Right? The will of the collective will override any one bad actor. So we have to educate ourselves, be more aware. One resource for this is my YouTube channel, but the internet is your university. So it's your responsibility to use it as such. I'm gonna end this talk with a wrap about decentralized AI. So I want everybody to stand up because I know you guys are sleepy. I know there's been some sleepy speakers. So please stand up everybody. All right? Everybody stand up. It's a, it's a wrap. It's called Smarter. It's a play on words. It's a parody of Kanye West Stronger. So I want you to put your hands up in the air as well. So let's just go. Let's go. I got to read papers to try and make me smarter. I trained my models in the cloud now because my laptop takes longer. I parsed through data like a boss now. Back then my code was wronger. Subscribe if you want to learn now. Let's spread this AI power. Let's train it right now. Let's train it right. Let's make bots tonight and make music like Mozart tonight and paint our work on a cross tonight. I've waited for this software all my life. Just give it some data and watch it explore. Step back clap and give it on core. It trained on 40 cores. Is racer make GPUs anymore? A game bot started off aimless. But now it's like God in the matrix. My chat bot speaks once I train it. I worked so damn hard to make this. It'll generate words that it predicts. Analyze sentiment then test it. This deaf is the shit to invest in. New data, wizard train it, then test it. Come in. I got to read papers to try and make me smarter. I trained my models in the cloud now because my laptop takes longer. I parsed through data like a boss now. Back then my code was wronger. Subscribe if you want to learn now. Let's spread this. All right everybody, thanks so much.